Pronto History Tour
Anthony Cassandra
Former CTO, Pronto LLC
Outline
- 2004 - An Idea
- 2005 - The Beginning
- 2006 - Transition to CSE
- 2007 - Social Shopping
- 2008 - SEM and Profitability
- 2009 - Microsite Stategy
- 2010 - Brain Drain
- 2011 - International SEM
- 2012 - Decline of SEM
- 2013 - The End
N/A
Goal: Downloadable Client
- Price comparison browser add-on (downloadable client)
- Alerts (a.k.a., "toasts") when viewing products on retail sites
- Price comparison vs. similar product alerts
- Don't call them "pop-ups"
N/A
The Corporation
- Name from United's "Red Carpet" lounges
- Incorporated March 17, 2005
- 80% IAC Owned
- Day 1 Founding Team: Dan Marriott, Brad Perry, Nancy Perry, Jonathan Marcus, Anthony Cassandra, John Foley, Sid Conklin
- Offices in Boulder, CO and New York, NY
- "Stealth" mode operation for 6 months
N/A
Early Data Challenges
- Crawling and extraction daily:
- of 60K+ web sites
- for 100+ million product offers
- Normalization for price comparisons
- Similar products logic
- Product web page modeling (for client)
- Implementing PERL regexes in Microsoft C++ world
N/A
Data Solutions
- Site models for extraction (xpath + regex)
- Page models for browser plugin (xpath + regex)
- Automatic model learning using domain-specific hints
- Product Classification (naive Bayes)
- Feature Extraction (context-sensitive rules)
- Normalization rules by features (fairly aggressive)
N/A
Pronto It Is
- Corporate name finalists:
- Pronto (so-so)
- Sweet (ugh)
- Uber (yuck!)
- Domain availability and cost limits the choice
- Beta launch in November, 2005
- First revenue event in late December, 2005
N/A
Transition to Comparison Shopping Engine
- Download-leary era, distribution never came
- Building a traditional comparison shopping engine (CSE)
- Revenue-sharing (CPA) gives way to cost-per click (CPC)
- Data feeds increase: direct and aggregate
- Initial friction between paid vs. unpaid listings
N/A
New Data Challenges
- Renewed focus on classification: browse taxonomy
- Classification peaked at 95% accuracy
- On average, ever page has an error
- Same offer, multiple sources (crawl vs. feed)
- Product search infrastructure not scaling
N/A
New Data Solutions
- Multiple classifiers: mixture of experts model
- Normalization: strong commodity, weak commodity, "titled"
- URL resolution to match products
- Unionization to merge "best" data from multiple sources
- Separation of data processing and user searches (data snapshots)
N/A
Pronto Web Site (version 1)
Persistent Data Challenges
- Feeds are not pristine data sources
- No standardization for feed syntax or product concepts
- Balancing data quality and economics
- Challenge to handle data error reports when known error rate exists
- Data error handling vs. source code error handling
- Simple Data Verification Tools w/Data Verfiiers
- Early Solution: Pronto Bug Firefox Add-on
N/A
Socially Transmitted Data Challenges
- Merging real-time user data with feed and crawl data
- Desiging to scale to millions of read/write transactions
- Notion of dynamic index an issue ("stale" products)
N/A
Pronto Web Site (version 2)
What Happened?
- Explosive traffic growth did not emerge
- Revenue pressure (still not profitable)
- Did not invest needed time and resources
- Should have tapped into Facebook network
N/A
Data Quality Advancements
- Data Doctor Platform
- Manual corrections, hints and feedback to data processing
- Company-wide competitions
N/A
2008 - SEM and Profitability
Profitability and SEM
- Meanwhile, SEM optimization engine was being refined
- SEM management tools being expanded and refined
- SEM drove Pronto to profitability in June, 2008
- More SEM with flanker brand OhDeal.com
- Improving SEM for Gifts.com (acquisition)
N/A
Economic-driven Data Challenges
- Merchant diversity in search becomes important
- Increasing friction: paid feed vs. unpaid crawl
- Data processing becomes more connected to business systems
- Testing data/logic changes increasingly challenging
- A/B testing for SEM becomes dominant force
N/A
Other Projects
- Flexible Search Filters
- Flexible Browse Taxonomy
- Pronto Newsletter
- Pronto Bookmarklet
N/A
Data Management Advancements
- GUI editors for data instead of text files
- Better version control and rollback abilities
- One button deployments and deployment histories
N/A
Fringe Projects
- Partner API Enhancements
(full site support)
- Pronto Labs Experiments
- New Shopping Helper Firefox Add-on
N/A
2009 - Microsite Strategy
Data Solutions for Verticals
- Birth of our "attribute crawl"
(really just page scaping)
- Supplemental data, merged with feed data
- Javascript and Flash sites particularly challenging
- Multiple browse taxonomy support
- Extracting features from images
(dress colors only)
N/A
Pronto Web Site (version 3)
IAC and Liberty Deal
- Frenemies: Diller and Malone
- Stocks, money and companies exchanged
- Gifts.com moves out of Pronto
N/A
The Brain Drain
- Pronto equity event and valuation
(Now 100% IAC owned)
- Many key technical people moved on
- Non-trivial systems with years of tweaks
- Frenetic startup activity made transition challenging
- "Best" software practices vs. "best" for business?
N/A
World-wide Data Challenges
- Data model and mechanisms had core language assumptions
- Separation of data for multiple data pipelines
- Language expertise hired: Japanese, German, French
N/A
Other Projects
- New iPhone application with price alerts (contract)
- New mobile web site
- Javascript Widgets (via Partner API)
N/A
SEM's Last Hurrah
- FineComb: Broader SEM via Google API
- AllBlox: Local SEM via Yellowbook API
- SEM risks leads to new business strategy
N/A
Re-focus and Re-organize
- MindSpark-driven revisiting of browser add-ons
- New editorial content strategy for direct traffic
- Website re-design
- Technical re-design
(service oriented architecture - SOA)
- Improving processes and project management
N/A
Casualties of Re-focusing
Web Site Redesign (version 4)
Fun Before the Storm
- First ever all company gathering in NYC (June 2012)
N/A
Lessons
- We did a lot with very little
- We did too much, not enough focus
- Short-term financial incentives detrimental
- Only ever had a single business metric
- "Best" software practices not always best
- Good people help compensate for flaws
- A lot of learning on the job
- It's been challenging, interesting and fun
N/A
